XPENG RoboticsMay 2021 – Sep 2021

3D Object Detection on Indoor Environments

An indoor 3D perception pipeline combining stereo depth, 3D object detection, and 6D pose estimation.

Project Snapshot

Signal Details
Timeline May 2021 to Sep 2021
Organization XPENG Robotics
Focus 3D Object Detection on Indoor Environments
Stack Custom implementation and project-specific tooling

Pipeline

  1. Generate accurate point clouds using stereo-based depth estimation.
  2. Run 3D object detection on the generated indoor point clouds.
  3. Use 6D pose estimation and PnP to recover more accurate dimensions and poses for smaller objects.
  4. Fuse 3D object detection with 6D pose estimation for richer indoor object understanding.

Why Fusion Matters

Sparse point clouds are effective for larger objects, but smaller objects may not have enough points for reliable 3D detection. The 6D pose branch helps recover details that pure point-cloud detection can miss.